Machine learning applications in colorectal cancer prediction

Colorectal cancer (CRC) is the third most commonly diagnosed malignancy and the fourth leading cause of cancer death in the world. Thus, its early detection through screening and removal of adenomatous polyps, prevents cancerous formation and reduces the incidence and mortality rates of this cancer. With the development of high technology in producing health informatics data, machinelearning applications can help us to improve the early detection of cancer. The aim of this study is to apply a new application of machine learning in order to do a classification between positive andnegative polyp/tumor groups on colonoscopy patients in Reza Radiotherapy and Oncology Center (RROC). In this study, we examined the demographic and clinical data of 693 patients (Positivepolyp/tumor=178 and Negative polyp/tumor=515) who underwent colonoscopy. Polyp and adenoma diagnosis were confirmed by pathology examination. Feature selection filter methods, such as theFisher and Gini index were employed in order to find the most informative features which play a key role in the categorization of the patient. Afterwards, the classification process is applied using the k-NNand SVM classifiers. The results of the proposed method based on the evaluation criteria such as accuracy, sensitivity, and specificity testified the validity of our method in comparison with the other classification methods on the CRC data. The best results from the current work on the CRC dataset using the k-NN and SVM classifiers shows an accuracy of more than 76.7 percent. By taking fulladvantage of filter methods for feature selection and k-NN and SVM classifiers, the proposed method is shown to be accurate and effective for early predictions of adenomatous polyps and colorectalcancer.

کلیدواژه ها:

Colon Cancer ، Colorectal Cancer ، Cancer Prevention ، Cell and Cancer ، Lifestyle and Cancer